1

The role of the distr ibution platform in price formation of paid apps

Paolo Roma

DICGIM 6 Management & Economics Research Group

Università degli Studi di Palermo

Viale delle Scienze, 90128, Palermo, Italy

paolo.roma@unipa.it

phone: +39 091 238 618 73

corresponding author

Fabio Zambuto

DICGIM 6 Management & Economics Research Group

Università degli Studi di Palermo

Viale delle Scienze, 90128, Palermo, Italy

fabio.zambuto@unipa.it

Giovanni Perrone

DICGIM 6 Management & Economics Research Group

Università degli Studi di Palermo

Viale delle Scienze, 90128, Palermo, Italy

giovanni.perrone@unipa.it

2

The role of the distr ibution platform in price formation of paid apps

Abstract

In this paper we study the role of the distribution platform as an important determinant of price of paid apps.

�!� �(/+� !4�)%*!�3$!0$!.� 0$!� ,.%�!� %),(%��0%+*/� +"� !2!(+,!./9� /0.�0!#%�� !�%/%+*/� �0� 0$!� �,,� (!2!(� $%*#!�

upon the targeted distribution platform. To these purposes, we construct a hierarchical model of price

formation by using an ad-hoc panel dataset consisting of top paid apps from the two major app stores,

*�)!(5� ,,(!9/� ,,��0+.!��* ��++#(!��(�5���1.�"%* %*#/�/$+3�0$�0�,.%�!/�+"�,�% ��,,/�/0.+*gly depend on

the platform where the apps are marketed. Specifically, the App Store is associated with lower prices for

paid apps than Google Play. We find evidence that this is because the impact of cross-store differences in in-

store developer competition prevails over the impact of cross-store differences in average consumer

willingness to pay. We also find that price premiums as a return to trialability are more likely to emerge in

Google Play than in the App Store, and that developers are more likely to adopt a penetration price strategy

in Google Play, arguably because the price implications of these decisions heavily rest on the willingness to

pay of consumers in the store. Finally, our evidence does not confirm the argument that a more marked price

reduction for paid apps operating as two-sided markets should be observed in Google Play.

K eywords: Mobile App Market, Online Distribution, Pricing, Electronic Commerce, Multi-level Data.

3

1 Introduction

In the very last few years, mobile software applications (apps, hereafter) have become important components

+"�,!+,(!9/�!2!.5 �5�(%"!�� /���)�00!.�+"�"��0��0$!�0%)!�,!.� �5�/,!*0��5�����+*/1)!./�1/%*#��,,/�%*�.!�/! �

by 35% in 2012, and was higher than the time spent on the web and quite close to that spent watching TV

(Khalaf 2012). Also, recent estimates suggest that the revenue generated from customers buying and

downloading apps for smartphones and tablets will step from $26 billion in 2013 up to $77 billion in 2017,

whereas the number of downloads will increase from 102 billion to 268 billion (Gartner 2013; Lunden

2013).

Although apps for mobile devices have been around since the late 90s, the app market started soaring in

2008 when Apple introduced the application store distribution paradigm. An application store is essentially

an online distribution platform (or e-marketplace) from which users can download apps developed by third

parties for mobile devices and operating systems (OS) supported by the platform. In response to the great

success obtained by Apple, several big mobile device manufacturers, such as Blackberry, Samsung and

Nokia, as well as software and Internet giants, such as Google, Microsoft and Amazon, have launched their

own app stores since 2008. However, as a result of growing market consolidation, two players undoubtedly

dominate the scene nowadays accounting for almost 90% of global app downloads (Gartner 2013)�� ,,(!9/�

App Store (with more than 1,400,000 apps, seventy-five billion cumulative downloads, and more than $10

billion cumulative store revenue) and Google Play (about 1,400,000 apps, fifty billion cumulative

downloads, and about $1.3 billion cumulative store revenue).1

While the rapid growth of the app market offers numerous business opportunities to a multitude of app

developers, it also provides a great opportunity for researchers to examine various theoretical issues like

innovation, entry and exit strategies, platform leadership, externalities, marketing mix (Boudreau 2012).

Initial studies have mostly concentrated on app demand estimation and the factors driving such demand

across major app stores (Carare 2012; Garg and Telang 2013; Ghose and Han 2014) or the determinants of

app success (Lee and Raghu 2014). In this paper, we instead study the role of the distribution platform in 1 Data on number of available apps, cumulative app downloads and revenue generated from apps were retrieved from the following sources: http://www.pocketgamer.biz/metrics/app-store/?mpage=appcount; http://www.appbrain.com/stats/number-of-android-apps; http://techcrunch.com/2014/06/02/itunes-app-store-now-has-1-2-million-apps-has-seen-75-billion-downloads-to-date/; http://www.techrepublic.com/article/google-play-v-apple-app-store-the-battle-for-the-mobile-app-market/; http://www.statista.com/statistics/281106/number-of-android-app-downloads-from-google-play/.

4

price formation of paid apps. Paid apps accounted for 75% of the revenue of app stores worldwide in 2013

and are expected to play a prominent role also in the future (Lunden 2013). In addition, industry figures

document that price levels of paid apps change significantly across stores, suggesting that pricing decisions

could depend on the distribution platform (Canalys 2012). Therefore, enhancing the understanding about the

impact of the distribution platform in price formation of paid apps is clearly important to developers who

have adopted paid app business model, as it can help them fine-tune their pricing decisions to the each

targeted platforms in order to better monetize on their apps. Elucidating the rationale behind the potential

emergence of differences in price levels across different online platforms is also particularly relevant for the

literature on online price formation. In fact, although price formation across online retailers has been

extensively studied in the past decade in numerous streams of literature including information systems and

economics (Brynjolfsson and Smith 2000; Clay et al. 2002; Chevalier and Goolsbee 2003; Baye et al. 2004a;

Chun and Kim 2005), the insights derived from these studies might not necessarily apply to app stores.

Indeed, there exist two fundamental differences, which distinguish app stores from traditional online retailers

and thus motivate our study on the role of the distribution platform in the app market.

First, online stores examined by previous studies are pure merchants who purchase from producers and resell

to customers by charging a retail price based on the wholesale price set by upstream firms (Hagiu 2007). For

this reason, the differences in price levels across stores documented by prior works in information systems

and economics literature are mostly explained by the different pricing and service policies experimented by

online retailers, thus implying the existence of a significant store effect in online price formation (Clay et al.

2002; Clemons et al. 2002; Baye et al. 2004b, Yao and Zhang 2012). Conversely, such practices are unlikely

to be the sources of price differences across stores in the context of apps, given that the prices to final

customers are set directly by developers, which then share the revenue for each unit sold with the platform

owner according to a fixed sharing rule, common to all major stores.

A second crucial difference lies in the unique relationship between the mobile OS/device market and the app

market. Once consumers choose their favorite OS/mobile device (e.g., Apple or Android), they are locked in

by this decision, as they rely exclusively on the sponsored platform to source their apps.2 This implies that

the customer base in a given platform strictly depends on the customer base of the associated product, i.e.,

2 Ghose and Han (2014) report that, at any given point in time, consumers usually own either Apple or Android devices.

5

the mobile device and the relative OS. Different mobile device makers and/or OS providers naturally target

different segments based on their product quality and marketing capabilities. In particular, with regard to the

two major mobile device ecosystems, it is largely recognized that Apple targets exclusively the (loyal) high-

end of the market, whereas the sales of Android devices are mostly fueled by low-end segments of the

market (Ghose and Han 2014; Nerney 2014). As a result, app developers face significantly different

consumer segments in terms of willingness to pay across different platforms (Forrester Research 2012;

Bensinger 2013; Edwards 2014), while this is unlikely to be the case in traditional online stores. This is

because in the latter stores consumers are not forced to bear any exclusive relationship with the distribution

channel, as there is no associated product and/or platform technology locking them in. Thus, the

heterogeneity in the willingness to pay of consumers buying the same item (e.g., the same electronic device

brand) across two traditional online retailers, e.g., Amazon.com and Newegg.com, is much lower than in the

case of consumers buying the same app in App Store and Google Play. This particular feature of the app

market implies that, differently from online retailers studied in previous literature, app stores can be viewed

as distinct markets where developers might be required to set different prices. Therefore, the role of the

distribution platform in price formation of paid apps should stem from driving forces, which are different

from those highlighted in previous studies, and thus, need to be unraveled.

Our paper aims at filling this important gap by studying the role of the distribution platform as an important

determinant of price of paid apps. We also add to prior literature on price formation in online markets by

examining whether the price implications of common app-level decisions made by developers are contingent

upon the targeted store; this being an issue that has been overlooked in such literature. Specifically, given

their dominant role in this market, we compare prices of top (i.e., most downloaded) paid apps from ,,(!9/�

App Store and Google Play and construct a hierarchical model of app price formation, taking into account

potential non-random selection issues by means of the propensity score matching (PSM) method

(Rosenbaum and Rubin 1983, 1984).

In detail, our contribution is twofold. First, we explain why the distribution platform should have a direct

influence on price levels of paid apps. We argue that the existence of significant heterogeneity in consumer

willingness to pay across different platforms should result in different platform attractiveness to developers.

The relatively low barriers to entry in this market imply fiercer developer competition in platforms accessed

6

by more valuable customers. As a result, we show that the direct impact of the distribution platform on prices

emerges from two contrasting forces: the average willingness to pay of consumers in the store, which favors

higher prices for paid apps in the App Store, and the induced in-store developer competition, which instead

favor higher prices for paid apps in Google Play. Second, we argue and find that cross-store heterogeneity in

consumer willingness to pay also influences the price implications of important app-level decisions that are

commonly made by developers to monetize on their paid apps. Specifically, we hypothesize and find that the

positive effect of offering a free trial version on the price of the full paid version is more likely to emerge in

platforms characterized by lower willingness to pay (i.e., Google Play), as less valuable consumers perceive

higher risks in buying an app of unknown value. We also hypothesize and find that, differently from the App

�0+.!�� 0$!� (+3!.� 3%((%*#*!//� 0+� ,�5� %*� �++#(!� �(�5� /$+1( � (%)%0� !2!(+,!./9� ��%(%05� 0+� .!/+.0� 0o price

skimming, and thus it leads them to adopt a penetration price strategy when launching new apps. However,

we do not find support for our hypothesis that the pricing of apps structured as two-sided platforms also rests

on the price sensitiveness of consumers.

The remainder of this paper is organized as follows. In § 2, we present our theoretical arguments and discuss

the related hypotheses. In § 3, we describe the data, the variables and the methods utilized in this paper. In §

4, we present our empirical findings and discuss their implications. In § 5, we discuss some robustness

checks. Finally, we conclude in § 6.

2 Theory and hypotheses

The central tenet of our arguments is that the distribution platform matters for price formation of paid apps in

two ways. First, we posit that the chosen distribution platform 6 App Store vs. Google Play 6 has a direct

impact on the price of a paid app, ensuing from the contrasting impact of two store-level forces, i.e., cross-

store differences in consumer willingness to pay and the implied extent of attracted in-store developer

competition. Second, we propose that the distribution platform also impacts prices indirectly, by influencing

0$!�,.%�!�%),(%��0%+*/�+"�0$.!!�%),+.0�*0� !2!(+,!./9� !�%/%+*/��0�0$!��,,�(!2!(�

2.1 The main effect of the distr ibution platform

7

In our view, cross-store differences in the average consumer willingness to pay are crucial in determining

price differences for paid apps across platforms.3 Intuitively, higher prices should be observed for paid apps

in platforms accessed by more valuable consumers (i.e. App Store rather than Google Play), as developers

can potentially extract higher value from them. However, cross-store differences in terms of consumer

willingness to pay also result in different platform attractiveness to developers. As mentioned, industry

evidence suggests that developers (especially those selling paid apps) generally find App Store more

�00.��0%2!�0$�*��++#(!��(�5���!��1/!� ,,(!9/��1/0+)!.���/!�%/�+*��2!.�#!�)+.!�2�(1��(e than Android-based

!2%�!/9��1/0+)!.���/!���+..!/0!.��!/!�.�$��������!*/%*#!.������� 3�. /�������%(�+4������� ��+. %*#�

to the industrial organization view of strategy (Porter 1980), a more attractive market also stimulates more

firms to entry to appropriate a slice of the bigger pie. In fact, when entry barriers are intrinsically low, a

highly attractive market might end up being characterized by fiercer competition than a less attractive market

because of the numerous firms that will actually entry and compete in the market (Siegfried and Evans 1994;

Lecocq and Demil 2006). This argument applied to the app market suggests that store level differences in

consumer willingness to pay could also determine significantly different levels of in-store developer

competition between the two platforms, namely App Store and Google Play.

Generally speaking, barriers to entry tend to be modest in the app market. This is evident considering that

traditionally important barriers to entry, namely capital requirement and product differentiation, play a

marginal role in the app market. As a matter of fact, development costs and the necessary process to market

most of the apps in major app stores can be afforded also by individual independent developers who can

launch their apps and compete with those of established software houses, even reaching top ranks. Also,

given the plethora of developers currently available in major app stores, product differentiation does not

appear to deter entries either. In addition to their modest magnitude in absolute terms, entry barriers tend be

3 One may argue that app distribution platforms also differ in several technical characteristics that might affect price. Specifically, Android OS is released under an open source license, as opposed to Apple iOS, which runs only for Apple devices. This may suggest that development costs could be higher in Google Play, due to the need for developers to customize apps for different smartphones. However, this should be balanced by the fact that app approval procedures are more stringent in the App Store, which implies higher quality control costs to ensure successful application when developing for this platform. Finally, there exist minute differences in the magnitude of the fees charged to developers to sell their apps. Accordingly, industry evidence suggests that development costs are overall comparable across the two platforms (Abrosimova 2014). More importantly, note that development costs are sunk investments rather than marginal costs, and thus should not have any bearing on pricing decisions. Finally, there are no differences in marginal costs between the two platforms as such costs are usually negligible in the app market. In fact, maintenance, scaling and user support costs are stepwise costs, and thus do not change in our short period of observation of prices. Overall, this implies that differences in the technical characteristics of the two stores (if any) are very unlikely to play a relevant role in cross-store differences in prices.

8

even less important in relative terms, as development costs in Apple are arguably comparable to those in

Google (Abrosimova 2014). Relying on these considerations, the higher willingness to pay of consumers

should attract more developers in the App Store as compared to Google Play, thus generating fiercer

competition in the former. Industry figures provide some support to this argument suggesting that developers

selling their apps in the App Store are almost twice as those in Google Play (Louis 2013). Moreover, this

effect becomes even stronger when considering paid apps. In fact, a larger number of developers prefer

selling paid apps in the App Store as a result of the higher willingness to pay of consumers in this store

(Wilcox 2014).

To sum up, cross-store differences in consumer willingness to pay determine the emergence of two

contrasting forces driving prices in opposite directions. On the one hand, the higher willingness to pay of

consumers in the App Store should offer developers the opportunity to charge higher prices for their paid

apps in this store as compared to Google Play. At the same time, the existence of more valuable consumers

(coupled with the modest entry barriers characterizing the app market) also has the effect of attracting

relatively fiercer developer competition in such platform as compared to Google Play, which, in turn, should

counteract the above effect by favoring higher prices for paid apps in Google Play through its negative

impact on price. As there is no theoretical argument a priori to anticipate which of the two effects actually

prevails, we formulate two alternative hypotheses on the overall net effect of the distribution platform, as

follows:

H1a. App Store is associated with higher prices than Google Play for paid apps, due to the higher average

consumer willingness to pay.

H1b. App Store is associated with lower prices than Google Play for paid apps, due to the higher level of in-

store developer competition.

2.2 Distr ibution platforms and the impact of developer decisions at the app level on app price

Developers can adopt a variety of strategies when selling their paid apps. They can offer limited free trials to

let consumer experiment with the product before committing to the purchase of a full paid version. In

addition, they can adopt different introductory pricing strategies for new apps (namely penetration price or

price skimming), or adopt a two-sided market model for their apps by relying on additional revenues e.g.,

revenue from ads inside their apps. While the price implications of these decisions have been extensively

9

studied for information goods (Gallaugher and Wang 2002; Shapiro and Varian 1999; Dubé et al. 2010; Liu

2010), little attention has been devoted to understand whether and how these implications are contingent on

the choice of the online distribution channel. More specifically, we argue that in the app market, it is

important for the chosen app-level decision to be aligned with the characteristics of the consumer segment

developers inherit from the associated device makers and/or OS providers, otherwise it may fail to produce

the expected consequences in terms of prices, and, ultimately, profitability. Therefore, in this section, we

explain th�0�0$!�%*"(1!*�!�+*�,.%�!�+"�,�% ��,,/�+"�0$.!!�%),+.0�*0� !2!(+,!./9� !�%/%+*/��0� 0$!��,,�(!2!(�6

i.e., offering a free trial version, structuring an app as a two-sided market, and choosing a pricing scheme for

new products 6 should be contingent on the chosen distribution platform.

F ree trial version

In general, trialability measures the extent to which potential adopters perceive that they have an opportunity

to experiment with an innovation prior to committing to its usage (Rogers 1983; Moore and Benbasat 1991;

Agarwal and Prasad 1997). Through a trial version users can test the product and resolve the uncertainty

about its real value to them (Wei and Nault 2013). Also, trialability can serve to signal product quality as the

knowledge that a product is available in a free trial version represents some sort of guarantee to customers.

For these reasons, consumers generally pay higher price if given the opportunity to test the product before

buying, consistently with the idea that they place value on reduced uncertainty surrounding the product

(Gallaugher and Wang 2002).

Developers often release a free trial version associated with the paid version of their apps. Therefore, the app

market is suitable to test the existence of a positive effect of trialability on the price of the paid version.

Nonetheless, while we take into account the impact of free trial versions on app prices in our analysis, our

main interest lies in testing whether the price implications of offering a free trial version depend on the

distribution platform where the app is marketed. In this respect, we argue that consumers with different

willingness to pay may have a different perception about the uncertainty surrounding an app, and thus they

could react differently to price strategies of developers. A consumer with limited willingness to pay could

perceive a substantial risk in buying an app of unknown value, despite the rather small price. Indeed, given

the limited budget to spend on apps, even wasting a couple of dollars on the wrong app could have serious

consequences because it might preclude the opportunity to increase device utility with alternative purchases.

10

Conversely, a more valuable consumer perceives less risk in buying an app of uncertain value, simply

because with its ample budget she can more easily substitute that app by purchasing better ones in case of

dissatisfaction. Therefore, consumers should be more likely to recognize price premiums for the full paid

version to the developers that help them resolve such uncertainty by releasing a free trial version, especially

in presence of lower willingness to pay.

As industry evidence has clearly shown that consumers have more limited willingness to pay in Google Play

than in the App Store, the above considerations suggest that the positive effect of trialability on the price of

the full paid version is more likely to be observed in the former store. Hence, we formulate the following

hypothesis:

H2: The positive effect of trialability on the price of paid apps is contingent upon the distribution platform.

In particular, such an effect is more likely to emerge in Google Play.

Two-sided market model

A two-sided market is a market characterized by the existence of cross-side externalities between two groups

of users, as the demand from one group strongly depends on the demand from the other group, and vice

versa. In such type of markets (e.g., TV and other media, game consoles), a profitable pricing structure

would imply to target aggressively, i.e., subsidize, the side of users who are able to exert a larger positive

externality on the other side, and monetize from the latter (Rochet and Tirole 2003; Parker and Van Alstyne

2005; Armstrong 2006; Li et al. 2010). As a consequence, the price charged to the subsidy side would be

lower in presence of cross-side externalities than it would be in absence of cross-side externalities.

Depending on the specific revenue model adopted by the developer, apps can themselves operate as a two-

sided market attracting app users on one side and interested third parties on the other side. Interested third

parties could be advertisers or firms seeking information about consumers for specific business purposes. In

"��0�����+*/% !.��(!�*1)�!.�+"�,�% ��,,/��/1�$��/�#�)!/��)+*#�+0$!./���+*0�%*�0$%. �,�.0%!/9�� /���lthough

with minor intensity as compared to free apps). Also, developers of certain apps, such as social networking

applications, center their revenue model on building a large user base to attract info seekers, and thus earning

from selling non-personally identifiable information (Williams 2009). As app users naturally exert an

irresistible appeal to third parties, i.e., advertisers or info seekers, it follows that developers aim at

subsidizing the consumer side when structuring their apps as a two-sided market. Thus, from the perspective

11

of app users, if there exist strong cross-side network externalities, we should observe that, ceteris paribus,

paid apps operating as a two-sided market (e.g., ad-supported apps or apps collecting and selling data) have

lower prices than those that do not exhibit such characteristics, as developers would subsidy users to attract

third parties from which to monetize. In contrast, developers of apps that do not operate as a two-sided

market do not have such an incentive to reduce the price to users given that they cannot rely on the money

side.

While providing evidence of a two-sided market effect in the context of apps could be interesting on its own,

our main interest lies in examining whether the effect of a two-sided market model hinges upon the chosen

distribution platform. Our argument is that the magnitude of the price reduction necessary to attract a

sufficiently large consumer base heavily depends on the type of consumers buying that app. Indeed, the

extent of the subsidy offered also depends on the price sensitiveness of the subsidy side; that is, larger price

reductions should be offered when the subsidy side is more price-sensitive (Rochet and Tirole 2003; Parker

and Van Alstyne 2005; Eisenmann et al. 2006). As app stores feature significantly different consumers in

terms of willingness to pay on average, this argument suggests that when a developer decides to structure an

app as a two-sided market, the price implications of such decision should differ across app stores. In

particular, we should expect larger price reductions, if any, for consumers in Google Play than those in App

Store, given their relatively lower willingness to pay. Hence, we can formulate the following hypothesis:

H3: The negative effect of the two-sided market model on the price of paid apps is contingent upon the

distribution platform. In particular, such an effect is more likely to emerge in Google Play.

Introductory pricing strategy

How firms should vary the price over the product life cycle to maximize their profits has been extensively

studied in marketing literature (Tellis 1986; Mahajan et al. 1990; Noble and Gruca 1999; Spann et al. 2015).

When introducing a new product firms can balance between two extreme pricing schemes, namely market

skimming or market penetration (Dean 1950; Mahajan et al. 1990). Under a skimming strategy, firms

initially set a high price targeting high valuable customers. Then, as the product moves forward in its

lifecycle, its price is gradually decreased to attract less valuable segments. On the other hand, a penetration

strategy implies the use of a low price at first to capture a larger market base. Then, once the product is

established in the market, the price can be raised.

12

Apps reasonably belong to a product category subject to network externalities. Therefore, the existence of

considerable word-of-mouth and other network effects among users should induce developers to utilize a

penetration strategy to exploit the advantage of a large installed market base (Shapiro and Varian 1999; Dubé

et al. 2010; Liu 2010). In this case, apps first released should have a lower price than more mature apps,

ceteris paribus. However, numerous successful apps have unique features that cannot be easily imitated or

substituted. Furthermore, for many apps there may be users who are highly valuable. For instance, very

passionate gamers can certainly afford to purchase their preferred app right after the release at any price. In

this case developers could find it optimal to skim this segment and later decrease the price to target less

addicted users, due to the presence of high customer heterogeneity and product differentiation (Dean 1969;

Tellis 1986; Noble and Gruca 1999). Under the circumstances, apps first released should instead have a

higher price than apps already mature in the market, ceteris paribus.

The above considerations would suggest that in the app market both pricing strategies are equally plausible,

thus yielding no univocal prediction on the relationship between the age of the app and its price.

Nonetheless, we suggest that the pattern of the price of new apps could be better understood after taking into

account also the characteristics of consumers accessing the given distribution platform. We argue that in

Google Play there could be limited room for a price skimming strategy. Since price skimming aims at

capturing the high end of a market, it requires the existence of a group of highly valuable consumers who are

willing to pay significantly more than the others for the product. This market segment could be relatively

small or less developed in Google Play, where consumers generally tend to have lower willingness to pay as

compared to the App Store. On the other hand, price penetration could be more impactful in this platform, as

less valuable and more price-sensitive users (who are the vast majority in Google) would respond more

favorably to lower initial prices, thus facilitating the establishment of an ample consumer base. The situation

is different in the App Store. In this platform consumers are on average wealthier and more willing to pay

than those in Google Play. Thus, it is certainly possible for some developers to benefit from skimming high-

end segments by charging higher prices to initial adopters. At the same time, other developers could also

succeed in the implementation of a penetration strategy when their apps feature important network

externalities. Accordingly, in Google Play developers should be more likely to implement a penetration

strategy rather than a skimming strategy. Thus, we should observe that, ceteris paribus, younger paid apps

13

are on average priced lower than more mature paid apps as prices gradually increase over time under a

penetration price strategy. Conversely, since both penetration and skimming strategies could work

effectively in the App Store, it is difficult to predict the dominance of either strategy. Hence, we simply

formulate the following:

Hypothesis 4: Developers are more likely to use a price penetration strategy in Google Play than in the App

Store.

3 Data and methods

3.1 Data

To test the above hypotheses, we built an ad hoc dataset by collecting data of top (i.e., most downloaded)

paid apps for smartphones in the Italian version of App Store and Google Play. We recorded data from the

top paid app ranking publicly available in each of these platforms on a daily basis in the period going from

March 7th to May 15th, 2013 (60 observation periods in total).

Top app rankings have been utilized by prior research analyzing the app market (e.g., Carare 2012; Garg and

Telang 2013). Similarly to Carare (2012), we restricted our focus to the top 100 apps. There are several

important reasons for why all studies consider top app rankings. First, these rankings are easily available

from the app stores as well as third parties, such as Distimo. Second, the insights obtained from studying

successful apps, rather than average apps, can be certainly more useful to developers that are planning the

development and marketing of new apps. Third and most important, although both App Store and Google

Play count more than one million applications available for download, the actual number of apps that can be

displayed by consumers is much more limited. In both stores consumers have access only to web pages

displaying top rankings (e.g., top free, top paid, top grossing) for all apps, top rankings within each app

category or top new entries and sponsored apps. Essentially only the very top portion of the app market is

actually visible to consumers. This implies that top rankings are arguably the primary source of information

not only for researchers to study this novel market, but also for consumers to make their purchase decisions,

as highlighted also by Carare (2012). Furthermore, besides the absolute relevance of paid apps (Lunden

2013), we naturally restrict to the top paid ranking and avoid considering top free apps (i.e., the most

downloaded free apps), given our focus on understanding the role of the distribution platform in price

formation of app prices. In this respect, we are consistent with previous econometric analyses on price

14

formation, which considered non-zero prices (Brynjolfsson and Kemerer 1996). Similarly, top grossing

rankings were also excluded, given the presence of numerous free apps in this ranking. Finally, our choice of

observing apps in a time span of about two months is in line with previous studies (Carare 2012; Garg and

Telang 2013). After recording the top 100 paid apps from the two stores along the entire period of

observation we obtained a rich dataset containing 11,999 observations related to 567 apps (402 apps in App

Store and 165 app in Google Play, respectively), that have appeared at least once in the top paid ranking of

one of the two stores during the observation period.4 By construction, our sample is an unbalanced panel

dataset as more successful apps appear in the rankings more often than others.

3.2 Variables

Dependent variable

Our theoretical arguments look at the relationship between prices of paid apps and the distribution platform,

as well as the role of the distribution platform in influencing the price implications of three strategies

adopted to monetize on paid apps. Hence, we consider the logarithm of price as our dependent variable and

record, on a daily basis, the price (in Euros) of each app featured in the top 100 paid apps rankings of the two

platforms during our period of observation.

Independent variables

Our hypotheses H1a and H1b suggest that the distribution platform chosen for a paid app will influence its

price. H1a suggests that the App Store will be associated with higher prices for paid apps than Google Play

due to the higher willingness to pay of App Store consumers, whereas H2b predicts the opposite outcome, as

more valuable consumers in the App Store will also attract relatively higher level of in-store competition

among developers in this platform as compared to Google Play. To test these predictions we first introduce a

dummy variable, namely Store, which is equal to one if on a given day an app is observed in the top 100 paid

app ranking of Google Play, zero if observed in the top 100 paid app ranking of App Store. Based on the sign

of the reported coefficient on our dummy Store we can assess whether one of the two opposing effects we

have theorized dominates the other, giving rise to significant differences in prices across stores.

4 The number of observations is 11,999 instead of 12,000 because we were not able to retrieve accurate and extensive information about one app appearing only once in the top ranking of Google Play. Moreover, the number of distinct apps is actually 524. This is because there are 43 apps available in the top 100 paid apps rankings of both stores during the observation period. Nevertheless, as explained later, according to our hierarchical model of price formation we consider the app-store pairs, which are indeed 567, as our statistical units.

15

At this point, it is important to recall that our first set of hypotheses looks at the overall net effect of the

distribution platform on the price of paid apps. Strictly speaking, if the two opposing store-level effects we

propose completely offset each other by driving prices in opposite directions, then our dummy Store may not

have any significant impact on price. Yet, the proposed mechanisms underlying H1a-H1b could be still both

at work in determining prices of paid apps. Thus, besides providing evidence that prices of paid apps

significantly differ between Apple and Google, we also provide further evidence to show that the differences

captured by our store dummy can be ascribed to the two underlying forces we propose 6 i.e., store level

willingness to pay and in-store competition. Ideally, we would like to replace the store dummy with two

(store-level) variables measuring more explicitly the average levels of consumer willingness to pay and

developer competition in the store, with the expectation that they should be different across the two stores

while having also an opposite impact on prices. However, while this is certainly doable for the competition

variable, it is more troubling to measure store-level consumer willingness to pay. Differences in levels of in-

/0+.!� !2!(+,!.��+),!0%0%+*���*�2�.5�+2!.�0%)!��!��1/!� !2!(+,!./9�!*0.%!/���* �!4%0/���.!�2!.5� 5*�)%��%*�

this market, therefore making it possible to measure the impact of changes in store-level developer

competition on app prices. The same approach cannot be used to measure store-level willingness to pay of

consumers simply because this factor is very stable over time (especially as compared to app price), as it

descends from strategic choices related to the type of market segments targeted by the associated device

and/or OS providers. As a matter of fact, using secondary data on consumer willingness to pay for a larger

time span (i.e., 2011-2014), we always observe a large and persistent gap in willingness to pay between

Apple and Google users. Thus, even in case of a larger time span, any measure of store-level willingness to

pay has almost zero variation in the sample and is highly correlated with the dummy Store, thus making it

difficult to discern its true effect on prices from the overall net effect.

Considering the difficulties in isolating the direct effect of cross-store differences in consumer willingness to

pay by explicitly measuring it, our approach to further shed light on how the distribution platform matters for

prices of paid apps is twofold. First, we show that there exist significant differences in the level of in-store

developer competition between the two platforms and then explicitly introduce in our models a store-level

developer competition variable under the expectation that it should have a negative impact on prices.

Second, to demonstrate that the store-level consumer willingness to pay can actually have the intuitive direct

16

impact we have theorized, we provide additional evidence that higher prices are indeed associated with App

Store than Google Play when considering only low valuable apps (defined as those belonging to the lowest

price quartile within each of the two app rankings) in both stores. Indeed, low valuable apps should

necessarily attract a larger number of developers in Google Play, given the lower average willingness to pay

of consumers accessing this platform. Accordingly, the differences in the level of developer competition

between the two stores should arguably be flattened when considering low-end apps, thus implying a

weakened effect of cross-store differences in competition on prices. As a result, the positive effect of store

level willingness to pay should emerge more strongly in this case.

Based on the above discussion, we can only use a measure of store-level competition in addition to the

dummy Store. However, as the number of sales or downloads of each developer in a given store is not

disclosed, we cannot compute measures of competition based on market shares. In addition, as the vast

majority of apps and developers are never displayed on the pages of the two stores, it is not possible to

retrieve time-varying information on the number of all developers in a platform and the number of their apps,

thus impeding the calculation of a Herfindahl-Hirschman Index (HHI) based on these data. To overcome this

problem, we measure the extent of developer turnover in the top paid apps ranking of the store. In fact, it is

straightforward that a larger turnover in the daily composition of top 100 ranking (in terms of developers and

apps) reflects fiercer competition among developers to increase download as well as higher intensity of new

market entries. Note that this choice is consistent with a consolidated stream in the industrial organization

literature using me�/1.!/� +"� "%.)/9� 01.*+2!.� %*� 0$!�)�.'!0� 0+� .!"(!�0� 0$!� (!2!(� +"� �+),!0%0%+*� ��5)!.� �* �

Pashigian 1962; Baldwin and Gorecki 1998; Caves 1998). To construct our measure of in-store competition

(namely, Turnover) we compute the number of developers that never appeared in the top ranking of paid

apps during our observation period over the seven days before the given day of observation.5 It is important

to note that our measure of competition is even more accurate than those that could be ideally computed

based on the number of apps of all developers in the store (if it were possible to compute them precisely).

This is because, with a mul0%01 !�+"��,,/�7%*2%/%�(!8�0+��+*/1)!./��0$!�)�&+.%05�+"� !2!(+,!./�+"�/1�$��,,/�

5 In this respect, it is important to note that looking at the turnover over the past week helps mitigate a potential endogeneity problem due to reverse causality between observed prices and our store competition variable. Also, note that our results are robust even when considering the new entries of developer in the top 100 paid apps ranking during a different number of days (e.g., 5 or 10) before the given day of observation.

17

are never a threat to developers of top apps, and a measure based on such information would not capture the

real level of competition among developers.6

To test our hypothesis H2, which suggests that the positive impact of trialability on the price of a paid app is

more likely to emerge in Google Play than in App Store, we retrieve information on the existence of a free

trial version for the app from the descriptions provided by app developers in the given store, and introduce a

dummy (Free trial version) indicating whether the developer offers a free trial version for the paid app or

not, as well as its interaction with our Store dummy.7 In our hypothesis H3 we maintain that a two-sided

market effect on price should be especially evident in Google Play given the lower willingness to pay of

consumers accessing this store. To test this hypothesis, we analyzed each app to check for the presence of

ads inside the app, as we had no access to revenue information from third parties, e.g. advertisers. In

addition, we considered all apps belonging to the social networking app category as apps operating as two-

sided markets. This is because the business model of these apps (e.g., Whatsapp) is explicitly centered on

gathering and selling (non-personally identifiable) data of a large user base to interested third parties.

Accordingly, we introduce a dummy variable (Two-sided market) equal to one if the given app contains ads

inside and/or belongs to the social networking category, as well as its interaction with the dummy Store. To

test our hypothesis H4, which suggests that developers are more likely to adopt a penetration price strategy

in Google Play than in App Store, we recorded the release date of the app in the given store. This allowed us

to compute the time (number of days) since the app has been available for download in the given store (Time

since release), i.e., the app age, and interact this variable with the dummy Store. Indeed, if developers in a

store tend to adopt on average a penetration (skimming) strategy, this measure should have a positive

(negative) impact on prices indicating that, as apps move forward in their lifecycle, developers tend to

increase (decrease) their price.8

6 In spite of the mentioned weaknesses of using data on the number of developers and their apps, we found that our results are robust also when using a daily HHI obtained by computing the market shares as the number of apps (both paid and free) of each developer featured in the top paid ranking of the given store on the given day divided by the total number of apps of all the developers featured in the top paid ranking of the same store on the same day. 7 Note that we distinguish between proper free trial version, i.e., version offering very limited features or being time-locked, and free version, i.e. version that does not differ much from the paid version except for the presence of (more) ads. Therefore, we include a specific dummy (Free version) to control for the effect of the latter type of version, which is not related to trialability. 8 An important caveat related to testing our proposed interactions (i.e., H2-H4) is that our dummy Store embodies two store-level effects. Thus, although the %/0.%�10%+*�,(�0"+.)��+1( � %*"(1!*�!� 0$!�,.%�!� %),(%��0%+*/�+"� 0$!� !2!(+,!./9�strategies discussed in H2-H4, it might be argued such influence is due to differences in in-store competition rather than to consumer willingness to pay as we propose. In order to rule out this alternative explanation we also explicitly

18

Finally, in addition to the variables of interest, we control for a number of other factors, which may influence

app price formation. First, in addition to the store-level competition variable, we introduce a variable

measuring competition at the app level in a given store. Indeed, different apps could have a different number

of substitutes marketed in the given store. To compute this measure, we considered all the apps marketed by

the developers observed in our sample. Based on this population, we computed for each app the number of

(both paid and free) substitutes at a given time (Substitutes). Note that by doing so, we did not only consider

the apps appearing in top 100 paid app rankings, but also those apps not appearing in the ranking, as long as

they were developed and marketed by the developers present in our sample. This is consistent with our

earlier argument that the major threats for an app likely come from developers that have been able to market

their apps more successfully, as most of developers and apps are never visible to consumers in both app

stores. To identify app substitutes more accurately we created 69 subcategories based on the real scope and

functionalities of the given apps retrieved from the descriptions provided by developers. Also note that, by

including free apps in computing the variable Substitutes, we explicitly capture the effect of the presence of

free apps on the price of paid apps. We also control for the app category (we count 14 categories in our

sample as indicated in Table 1), the type of developer (Developer type), i.e., whether the content provider is

a firm or (group of) individual(s), the number of apps marketed by each observed developer in the given day

and store (Number developer apps), and the app size in Megabyte (App size). We also control for whether in-

app purchase (i.e., the opportunity to directly upgrade to additional features inside the app) is implemented

for the given paid app, by introducing a dummy variable (In-app purchase). Furthermore, given that users

can rate the apps they download, we recorded such information for each app to construct a measure of app

-1�(%05���,!�%"%��((5���+0$� ,,��0+.!��* ��++#(!��(�5��((+3�1/!./�0+�,.+2% !��,,/9�.�0%*#�+*����-to-5 scale

where 1 corresponds to the worst valuation and 5 to the best valuation. However, sometimes no rating is

displayed for certain apps because the number of users who have provided a rating is too low. To cope with

this issue, we construct four dummies based on the rating: low rated apps (Low app rating) category if the

app rating is below 2.5, medium rated apps (Medium app rating) category if the app rating is between 2.5

and 3.5, high rated apps (High app rating) category if the app rating is above 3.5, and finally a category of

interacted the app pricing strategies variables with the Turnover variable. Our results showed that these interactions are never significant, and thus the difference in in-store competition is unlikely to be the driving forces behind H2-H4.

19

apps displaying no rating (No app rating).9 We also control for the extent of developer notoriety by including

a dummy (Developer fame), which indicates whether the developer/brand of the given app is worldwide

established.10

Table 1 shows the descriptive statistics of the variables employed in our study. On average, prices of paid

apps in Google Play are considerably higher than those in the App Store (1.87 and 3.41, respectively), and

both stores display significant within-store price variation. At the same time, observed competition levels are

higher in the App Store than in Google Play both at the store level (the variable Turnover is 25 in the App

Store and only 5.94 in Google Play) and the app level (the variable Substitutes is 687 in the App Store and

349 in Google Play). These figures seem to suggest that the different levels of in-store competition between

the two stores could play a significant role in determining prices of paid apps. As for our categorical

variables, Table 1 clearly suggests that the top paid apps rankings of the two stores do not display the same

composition in terms of apps and, thus, price differences across the two platform might be due to the

different apps featured in the rankings rather than to the arguments we propose. We will carefully address

this issue in the subsequent sections when discussing how matching similar apps across stores helps mitigate

these potential biases. Table 2 reports the correlation matrix. First, note that the dummy Store and the store-

level measure of competition are extremely correlated (Pearson correlation coefficient equal to -0.92), which

provides a first hint that there exist strong cross-store differences in the level of in-store developer

competition. At any rate, these two variables are not introduced at the same time in the regression analyses

that follow. Also note that, in such analyses, we choose the dummy Games and Medium app rating as

baselines for app category and app rating variables, respectively.

3.3 Methods

Our unbalanced panel dataset naturally displays a multi-level structure as each observation is related to the

daily price of a given app, marketed by a given developer, in a given store. Therefore, observations related to

the same app in the given store, as well as those pertaining to same developer in the given store are likely to

9 Note that there might be endogeneity due to potential reverse causality between price and app rating. In the robustness section, with the support of instrumental variables (IV) regression, we show this is not the case in our sample of apps. 10 Based on corporate revenue information and worldwide brand recognition we identify 22 top developers in our sample: Activision Blizzard, Adobe Systems, Apple, Autodesk, AVG Technologies, Capcom, De Agostini, Disney Interactive, Electronic Arts, Fox Digital Entertainment, Gameloft, Google, Konami, Mediaset, Namco Bandai, Research in Motion (Blackberry), Sega Sammy, Square Enix, Take-Two Interactive, TomTom International BV, Ubisoft Entertainment, Zynga.

20

be correlated. Similarly, observations in the same store could be correlated as they are exposed to common

store-level factors. Therefore, we propose a four-level model to analyze the role the distribution platform in

price formation of paid apps. The first level is the observation (i.e., the app in the given store on a given

day); the second level is the app-store pair (i.e., the app in the given store), the third level is the developer-

store pair (i.e., the developer in the given store), the fourth level is the store. Essentially, as app stores can be

viewed as different markets, we structure our dataset so that observations are nested within app-store pairs,

which are nested in developer-store pairs, which in turn are nested within stores. In presence of such multi-

level data structure, the use of mixed linear effects regression models is usually suggested in literature

(Raudenbush and Bryk 2002). As we are interested in shedding light explicitly on the role of the store (Apple

vs. Google) in price formation of paid apps, we introduce the distribution platform variable, the relative

interaction terms and control variables at both app and developers levels as fixed effects, while treating as

random the effect of the app-store and developer-store pairs. Hence, our full linear mixed effects model is as

follows:

�� �����-*+, / $& . $' � ����� . 0$( . $) � �����1 � ���������� ������-*+, .

����������0$� . $� � �����1 � �!���������������� . 0$� . $� � �����1 � ���������������������� .� (1)

" � �����������-*+, . # � �� ���������������-+, . �+, . �*+, . %-*+,� where, rijs and ujs are the app-store and developer-store specific random effects respectively, �tijs is the error

term, whereas the remaining terms are our variables of interests and controls modeled as fixed effects.

Another important concern in assessing the impact of the distribution platform on prices of paid apps is

related to the fact that consumers self-select in any of the two stores based on the type of device they

possess. This suggests that, in addition to the different willingness to pay, which is our object of study,

consumers could have substantially different tastes and needs across stores. As a consequence, developers

might strategically self-select in any of the two stores to offer apps that better match the tastes and needs of

consumers in that market, implying that apps observed in the top ranking of App Store could be

systematically different from those observed for Google Play. This might generate significant selection bias,

possibly invalidating our findings. In fact, as apps in our sample are not randomly assigned to the two

platforms, cross-store differences in prices could be simply due to the different types of paid apps featured in

the two stores, rather than to the two driving forces we have theorized. Ideally, considering exactly the same

apps in the two stores would arguably solve this issue. However, only 43 apps out of 567 apps are available

21

in the top 100 paid apps rankings of both stores during the observation period, thus not allowing a reliable

econometric analysis. Therefore, in order to mitigate this selection problem and ensure a comparison across

stores for relatively similar apps, we apply the Propensity Score Matching (PSM) method (Rosenbaum and

Rubin 1983, 1984). To avoid the biases that nonrandom assignment to treatment may generate, matching

algorithms are commonly used to find a non-treated unit that is similar to a participating unit across several

dimensions.

We performed the matching based on a set of both categorical as well as integer/continuous variables, and

considered observations in App Store as treated units, while those in Google Play as non-treated units.11 We

applied PSM within strata, by considering exact matching for app categories (which identifies the app

functionalities) and app rating levels (a proxy for quality). For instance, we matched games having high

rating in App Store only with games having the same rating in Google Play, and so forth. Within each of

these strata, we applied the PSM and considered a set of integer/continuous covariates to further match

treated and non-treated apps. We used the variable Time since release, as price changes along the app life

cycle and stores could feature apps of different maturity. Similarly, we considered the daily rank of the app

to pair apps enjoying similar exposure to consumers in the two stores. After matching, our sample is reduced

to 3,520 observations of which half are from App Store and half from Google Play, due to the fact that we

apply one-to-one matching. By construction, these matched samples have the very same composition in

terms of app categories and rating level. Unreported t-tests and ANOVA analyses also confirm that matched

apps in the two stores do not differ significantly in terms of age and ranking (covariates used for the

matching), as well as in terms of app size. Hence, we are confident that performing the above matching helps

)%0%#�0!�,+0!*0%�(��%�/!/�%*�+1.�/�),(!��.%/%*#�".+)��+0$��+*/1)!./9��/�3!((��/� !2!(+,!./9�/!("-selection in

the two stores. Finally, note that for robustness purposes we created alternative matched samples by pairing

apps also based on additional covariates12, and implemented alternative matching procedures such as

11 We imposed a common support (overlap condition) with no-replacement by removing treatment observations for which the propensity score was higher than the maximum or less than the minimum of the score of non-treatment observations. We also used the caliper options to select one and only one non-treated unit for each treated unit only when the propensity score difference was at most 0.001 (i.e., caliper set equal to 0.001). 12 For our main analyses we did not consider the variable App size as one of the features utilized for the matching, because the app size differ across the two stores, even for the very same apps. To give the idea, the size of TomTom Italy navigator is 404 Mbyte in App Store, whereas it is only about 29 Mbyte in Google Play. Therefore, using the size might actually distort our matching algorithm to the extent that it would not even pair the very same apps. At any rate, as we will discuss later, we created several matched samples based also on the variable App size as well as other covariates, such as F ree trial version, In-app purchase, and Two-sided market, showing full robustness of our results.

22

coarsened exact matching. These additional analyses are fully consistent with those reported in the paper,

and are available from the authors upon request.

4 Results and discussion

Descriptive statistics in Table 1 show that on average paid apps have higher prices in Google Play than in the

App Store (3.41 versus 1.87). This difference seems inconsistent with the intuitive direct effect of store-level

consumer willingness to pay, which should favor higher prices for paid apps in the App Store. Still, this does

not guarantee that the lower prices in the App Store are the result of the fiercer developer competition in this

store. In fact, differences in mean prices might still be driven by the different app composition in the top paid

apps rankings resulting from the choices of consumers and developers that self-select in the two platforms.

We have pointed out that matching apps according to their basic features should reduce (if not eliminate)

such potential biases. Thus, before presenting our more formal regression models based on (1), we provide

some preliminary evidence of the effect of our matching procedure on our dependent variable, i.e., the app

price. T-test results reported in Table 3 show that significant differences (0.1% level of significance)

between the average prices of the two stores emerge even when comparing matched pair of similar apps,

although the magnitude of this difference is considerably reduced (from 1.54 to 0.53). Therefore, price

differences across the two stores cannot be explained exclusively by different app composition of the two

stores. Table 3 also reports the cross-store difference in the means of our two competition measures at both

store and app-store levels, i.e., Turnover and Substitutes, respectively. The results clearly show that, even

after matching similar apps, the two platforms differ significantly in terms of competition with the App Store

displaying considerably higher competition. Overall, this preliminary univariate analysis provides first

support to our general argument that the distribution platform could influence price formation of paid apps

and that the two major app distribution platforms differ not only in terms of consumers willingness to pay,

but also in terms of in-store developer competition.

To test our hypotheses more formally we conduct a number of regression models based on the mixed linear

model presented in (1). Given the potential presence of selection bias in the full sample, we run these models

for the matched sample and report the results in Table 4 (the results under the full sample are qualitatively

unchanged and available from the authors). With regard to our contrasting hypotheses (H1a and H1b) on the

overall effect of the distribution platform, the first column of Table 5 shows that Google Play is associated

23

with higher prices than App Store due to the positive (and significant) coefficient of the dummy Store

(�1=0.232 and p<0.001). First, this finding confirms that price levels of paid apps vary across the

distribution platforms. Second, it provides support to H1b against H1a, suggesting that the effect of the

lower in-store competition in Google Play prevails over the direct (and opposing) effect of the lower

consumer willingness to pay in this store, thus leading to higher prices in this platform than in the App Store.

To further increase confidence about this interpretation, we substitute the dummy Store with our competition

measure at the platform level, namely Turnover. The second column of Table 4 shows that the coefficient of

this variable is negative and strongly significant, thus substantially capturing the impact of the dummy Store.

This result combined with the highly significant gap in competition between the two platforms (shown in

Table 3) provides stronger support to the argument that the level of in-store competition is the major driving

force behind the overall effect of the distribution platform on prices, thus leading to higher prices in Google

Play for paid apps.

Although results in the first two columns support H1b, one could question the existence of a conflict between

the effects of the different willingness to pay of consumers in the two stores and the consequent different in-

store competition. In columns 3-4 we provide evidence that, while in-store competition seems to explain

overall price differences across stores, the willingness to pay does actually have a direct impact on price

levels of paid apps. As explained, since we are unable to measure such variable explicitly, as an alternative

approach we show its impact by focusing on a sample of low valuable paid apps. In presence of low valuable

paid apps the differences in the level of in-store developer competition between the two stores should

arguably be lessened, as developers in Google Play should be more likely to concentrate on marketing low

valuable apps given the lower average willingness to pay of consumers accessing this platform. In fact, our

data show that, while the average number of app substitutes in Google is roughly only 50% of that observed

in Apple in the full sample, this difference significantly shrinks for the subsample of low-end apps, for which

the number of substitutes in Google is almost 88% of that in Apple. Yet, while competition levels become

more similar, there may still persist differences in terms of willingness to pay of consumers accessing the

different platforms. Thus, if the role of the distribution platform actually emerges from the trade-off between

these two factors, the negative impact of the dummy Store should be mitigated in this subsample of apps or

eventually revert. Results in column 3 confirm this by showing that the coefficient of the dummy Store

24

becomes significant and negative when we run our mixed model for the subsample of low valuable apps. The

marginal role of competition in presence of low valuable apps is confirmed by observing that our measures

of competition at the app level and store level are never significant in columns 3-4. Thus, in the case of low-

end apps price differences across the two platforms seem to be explained more by different store-level

consumer willingness to pay rather than by different in-store developer competition. Overall, this first set of

evidences provides support for our H1b, and confirms that the impact of the distribution platform on app

prices emerges as the result of a trade-off between the effect of differences in consumer willingness to pay

and the implied differences in in-store competition.

Columns 4-8 of Table 4 report the results for our second set of hypotheses predicting how the distribution

,(�0"+.)�%*"(1!*�!/�0$!�.+(!�+"�/+)!�%),+.0�*0� !2!(+,!./9� !�%/%+*/��0� 0$!��,,�(!2!(�%*�,.%�!�"+.)�0%+*�+"�

paid apps. Specifically, in the fifth column, we add the interaction term between the dummy Store and the

dummy Free trial version to the model in column 1. First, note that in column 1 the coefficient of the

dummy Free trial version was not significant, suggesting that on average trialability does not imply any

price premium for the full paid version. After introducing the interaction term, the coefficient of this dummy,

which now reflects the marginal change in price due to trialability in the App Store, is still not significant.

However, the coefficient of the interaction term, which instead reflects the marginal change in price in

Google Play due to trialability, is largely significant (at the 1% level) with a positive sign (�3 =0.305), thus

supporting our hypothesis H2, which suggests that the impact of a free trial version on the price of the full

paid version is more likely to emerge in Google Play. As we have argued, consumers in Google Play

perceive higher risks in buying an app of unknown value than consumers in the App Store due to their lower

willingness to pay. Thus, they seem to be willing to recognize a price premium to developers offering a free

trial version to resolve the uncertainty surrounding the value of the paid version.

In the sixth column we add the interaction term between the dummy Store and the dummy Two-sided market

to our baseline model shown in column 1 (where the latter was not significant). We find that the coefficients

of the two-sided market dummy and its interaction term with the distribution platform are both shown to be

not significant. Therefore, H3 is not confirmed in our sample, and apps operating as two-sided markets do

not seem to be associated with significantly different prices across the two platforms.

25

In the seventh column we add the interaction term between the dummy Store and the variable Time since

release to the baseline model in column 1. The coefficient on this variable was shown to be strongly

significant with a positive sign in column 1, which suggested that, on average, developers increase prices as

apps move forward in their lifecycle, consistently with the adoption of a penetration price strategy. However,

after adding the interaction term, the coefficient on the variable Time since release, which now reflects the

marginal change in price in the App Store due to the increase in app age, is no longer significant.

Conversely, the interaction of this variable with the distribution platform, which now reflects the potential

marginal change in price in Google Play due to the increase in the age of the app, is significant with a

positive sign (�7 =0.0004 and p<0.001). This result supports H4, by showing that price penetration is more

impactful in Google Play than in App Store, arguably because the segment of highly valuable consumers

could be relatively less developed in the former platform to allow the adoption of price skimming practice,

given their lower willingness to pay. The last column of Table 4 reports our full model including all three

interaction terms and shows full consistency with the results in the previous models.

At this point, it is important to recall that the results in Table 4 are based on a sub-sample obtained by

matching apps according their basic characteristics. Therefore, one could argue that the results on the

interactions between the distribution platform and app-level decisions might be still biased by the fact the

two rankings display different composition in terms of percentage of free trial version, app operating as two-

sided markets, and so forth. Nevertheless, even after refining our matching procedure by exactly matching

also for all app-level decisions variables (e.g., observations of apps offering a free trial version in App Store

matched only with observations of apps offering free trial version in Google Play, and so forth), our findings

remain unchanged in spite of a considerable reduction in the number of observations (from 3,520 to 1,520).13

5 Robustness checks and further discussion

In this section, we briefly discuss the results of further additional analyses we conducted to test the

robustness of our prior findings. The details of these robustness checks are available from the authors. First,

our results are qualitatively unchanged when eliminating those apps featured in the top paid app ranking

sporadically (e.g., less than three or four occurrences) to ensure comparison among apps of similar success.

Our second extension addresses potential endogeneity concerns related to time-varying app characteristics.

13 The analysis under this refined matching procedure is available from authors upon request.

26

Specifically, while in the models above we control for unobserved app quality by means of app rating, we do

not control for the potential effect of mobile app trend due to advertising or online word-of-mouth. As we do

not have access to explicit information about the advertising expenditures or the extent of word-of-mouth,

we utilize a proxy that reflects the mobile app trend over time. Specifically, we utilize the tool Google Trend

provided by Google, which provides a weekly index on a 0-100 scale of search popularity of words on

Google search engine in specific geographical regions. We recorded such weekly index during our period of

observation by searching the exact name of all the apps in our sample. Given the popularity of Google search

engine, this measure reflects the trend of an app in a given period, as a higher index is the result of greater

knowledge and interest about the product, which, in turn, is likely to be the consequence of more effective

advertising and online word-of-mouth over time. While being not significant, this new variable does not

change our main results. Our findings are also robust when we include one dummy for each day of

observation to account for the effect of trends common to both app stores and all observed apps.

Finally, endogeneity concerns might emerge also from reverse causality in the relationship between price

and app rating. In fact, the rating of an app might be affected by the price of the same app as a higher price

might create higher expectations of consumers, who will be more likely to complain about product flaws and

provide lower ratings to the given app. To address this concern, we perform instrumental variables (IV)

regression models, where we use developer rating as an instrument for app-level rating. This choice is

motivated by the idea that, on average, a developer marketing high quality apps is more likely to develop an

app of high quality than a developer that has been marketing low quality apps. We obtain the developer

rating in a store as the average rating of all the apps (except the considered app) marketed by the developer

in the given store until the given day. As both exogeneity test as well as the Sargan-Hansen

overidentification test (which evaluates the statistical validity of instruments) are consistently insignificant,

we can be sufficiently confident that our main findings are not affected by potential endogeneity concerns

due to the app rating variable. At any rate, the findings are robust even under the IV regression model.

6 Conclusions

In this paper, we have contributed to the literature on price formation in online markets by examining the

unique role of online distribution platform, specifically App Store vs. Google Play, in price formation of paid

apps. Differently from online retailers studied in previous literature, in app stores prices to final consumers

27

are set by developers and, more importantly, app consumers are locked-in by their decision to purchase a

device linked to an exclusive distribution platform. Accordingly, we have argued that differences in price

levels across different stores emerge as a result of the fact that developers face different consumer segments

and market conditions in different app stores, rather than heterogeneous pricing and service policies across

retailers.

In detail our findings are as follows. First, price levels of paid apps strongly depend on the targeted

distribution platform, and specifically on average Google Play is associated with higher prices than App

Store for paid apps. We have argued that the platform effect emerges as the result of two contrasting forces,

i.e., the consumer willingness to pay (which would imply higher prices in the App Store), and the in-store

developer competition (which would instead imply higher prices in Google Play). Our findings suggest that

in general differences in developer competition at the store level seem to be more influential in determining

different prices across the two stores than differences in average consumer willingness to pay. Further

confirming the existence of a trade-off, we find that price differences across stores are actually reversed for

the restricted subsample of less valuable apps. Second, our findings suggest that the distribution platform

heavily bears on the price implications of two important strategies adopted by developers. Specifically, we

find that price premiums on the full paid version of the app as a return to the provision of a free trial version

are more likely to be observed in Google Play, where consumers are more likely to suffer from information

asymmetry given their limited willingness to pay. Also, the distribution platform appears to be important for

pricing strategies used when launching new apps. In this respect, we find developers are more likely to use a

penetration price strategy in Google Play, whereas they possibly use both penetration price and price

skimming strategies in App Store. The rationale is that the segment of highly valuable consumers could be

relatively less developed in Google Play to allow the adoption of a profitable price skimming practice.

Finally, no univocal evidence supports the claim that adopting a two-sided market model should yield a

larger price reduction in Google Play than in App Store as a result of the lower willingness to pay of

consumers accessing this store.

These findings offer several remarkable implications for more informed decisions in the app market. A

primary implication of our findings is that app developers need to adjust their pricing decisions to the

targeted store, as different platforms imply consumers with different willingness to pay and, ultimately, also

28

different levels of developer competition. Also, they may need to make accurate app-level decisions,

especially those related to free trial version release and introductory price for new apps, according to the

targeted platform in order to better profit from their apps. Finally, our findings help increase distribution

,(�0"+.)�+3*!./9��3�.!*!//�+*�0$!�.+(!�+"�/0+.!-level factors in price formation and rethink their strategies to

attract both users and developers.

We expect that many researchers will devote themselves to the study of the app market given its growing

popularity. Accordingly, we believe that there is large room for further investigating the role of the

distribution platform. For instance, to complement our study, it would be interesting to investigate whether

developer choices related to app characteristics are influenced by the store choice, thus going beyond their

mere price implication as done here. Further research could also examine the factors guiding developers

entry in the platform choice when launching a new app. Specifically, this direction could provide answers to

questions such as whether to launch the new app only in one store, which store to target first, whether to

entry simultaneously, and the relative rationale and performance implications of such choices.

References

Abrosimova, K. 2014. How much does it cost to make iPhone, iPad and Android apps?. May 28. Accessed

June 11, 2015. http://yalantis.com/blog/iphone-ipad-android-app-development-cost/.

Agarwal, R., J. Prasad. 1997. The role of innovation characteristics and perceived voluntariness in the

acceptance of information technologies. Decision Sciences, 28(3), 557-582.

Armstrong, M. 2006. Competition in two-sided markets. RAND Journal of Economics, 37(3), 668-691.

Baldwin, J. R., P. Gorecki. 1998. The Dynamics of Industrial Competition: A North American Perspective.

Cambridge, UK: Cambridge University Press.

Baye, M. R., J. Morgan, P. Scholten. 2004a. Price dispersion in the small and in the large: Evidence from an

Internet price comparison site. The Journal of Industrial Economics, 52(4), 463-496.

Baye, M. R., J. Morgan, P. Scholten. 2004b. Temporal price dispersion: Evidence from an online consumer

electronics market. Journal of Interactive Marketing, 18(4), 101-11.

Bensinger, G. 2013. The evolving economics of the app. Wall Street Journal, March 3. Accessed June 11,

2015. http://www.wsj.com/articles/SB10001424127887323511804578300183961009070.

29

Boudreau, K. J. 2012. Let a thousand flowers bloom? An early look at large numbers of software app

developers and patterns of innovation. Organization Science, 23(5), 1409-1427.

Brynjolfsson, E., C. F. Kemerer. 1996. Network externalities in microcomputer software: An econometric

analysis of the spreadsheet market. Management Science, 42(12), 1627-1647.

Brynjolfsson, E., M. D. Smith. 2000. Frictionless commerce? A comparison of Internet and conventional

retailers. Management Science, 46(4), 563-585.

Canalys. 2012. Android apps are too expensive. February 23. Accessed June 11, 2015.

http://www.canalys.com/newsroom/android-apps-are-too-expensive.

Carare, O. 2012. The impact of bestseller rank on demand: Evidence from the app market. International

Economic Review, 53(3), 717-742.

Caves, R. E. 1998. Industrial organization and new findings on the turnover and mobility of firms. Journal of

Economic Literature, 36(4), 1947-1982.

Chevalier, J., A. Goolsbee. 2003. Measuring prices and price Competition online: Amazon.com and

BarnesandNoble.com. Quantitative Marketing and Economics, 1(2), 203-222.

Chun, S-H., J-C. Kim. 2005, Pricing strategies in B2C electronic commerce: Analytical and empirical

approaches. Decision Support Systems, 40(2), 375-388.

Clay, K., R. Krishnan, E. Wolff, D. Fernandes. 2002. Retail strategies on the web: Price and non-price

competition in the online book industry. The Journal of Industrial Economics, 50(3), 351-367.

Clemons, E. K., I. H. Hann, L. M. Hitt. 2002. Price dispersion and differentiation in online travel: An

empirical investigation. Management Science, 48(4), 534-549.

Dean, J. 1950. Pricing policies for new products. Harvard Business Review, 28(6), 45-53.

Dean J. 1969. Pricing pioneering products. The Journal of Industrial Economics, 17(3), 165-179.

Dubé, J-P. H., G. J. Hitsch, P. K. Chintagunta. 2010. Tipping and concentration in markets with indirect

network effects. Marketing Science, 29(2), 216-249.

Edwards, J. 2014. These maps show that Android is for people with less money. Business Insider, April 3.

Accessed June 11, 2015. http://www.businessinsider.com/android-is-for-poor-people-maps-2014-4?IR=T.

30

Eisenmann, T., G. Parker, M. W. Van Alstyne. 2006. Strategies for two-sided markets. Harvard Business

Review, 84(10), 1-11.

Forrester Research. 2012. Mobile apps for marketing. September 2012, Forrester Research Report,

www.forrester.com.

Gallaugher, J. M., Y. Wang. 2002. Understanding network effects in software markets: Evidence from web

server pricing. MIS Quarterly, 26(4), 303-327.

Garg, R., R. Telang. 2013. Inferring app demand from publicly available data. MIS Quarterly, 37(4), 1253-

1264.

Gartner. 2013. Gartner says mobile app stores will see annual downloads reach 102 billion in 2013.

September 19. Accessed June 11, 2015. http://www.gartner.com/newsroom/id/2592315.

Ghose, A., S. P. Han. 2014. Estimating demand for mobile applications in the new economy. Management

Science, 60(6), 1470-1488.

Hagiu, A. 2007. Merchant or two-sided platform?. Review of Network Economics, 6(2), 115-133.

Hymer, S., P. Pashigian. 1962. Turnover of firms as a measure of market behavior. The Review of

Economics and Statistics, 44(1), 82-87.

Khalaf, S. 2012. Mobile apps: We interrupt this broadcast. December 5. Accessed June 11, 2015.

http://blog.flurry.com/default.aspx?Tag=Television.

Lecocq, X., B. Demil. 2006. Strategizing industry structure: The case of open systems in a low-tech industry.

Strategic Management Journal, 27(9), 891-898.

Lee, G., T. S. Raghu. 2�����!0!.)%*�*0/�+"�)+�%(!� �,,/9� /1��!//���2% !*�!� ".+)� 0$!� ,,��0+.!�)�.'!0��

Journal of Management Information Systems, 31(2), 133-170.

Li, S., Y. Liu, S. Bandyopadhyay. 2010. Network effects in online two-sided market platforms: A research

note. Decision Support Systems, 49(2), 245-249.

Liu, H. 2010. Dynamics of pricing in the video game console market: Skimming or penetration?. Journal of

Marketing Research, 47(3), 428-443.

Louis, T. 2013. How much do average apps make?. Forbes. August, 10. Accessed June 11, 2015.

http://www.forbes.com/sites/tristanlouis/2013/08/10/how-much-do-average-apps-make/.

31

Lunden, I. 2013. Gartner 102B app store downloads globally in 2013, $26B in sales, 17% from in-app

purchases. September 19. Accessed June 11, 2015. http://techcrunch.com/2013/09/19/gartner-102b-app-

store-downloads-globally-in-2013-26b-in-sales-17-from-in-app-purchases/.

Mahajan, V., E. Muller, F. M. Bass. 1990. New product diffusion models in marketing: A review and

directions for research. Journal of Marketing, 54(1), 1-26.

Moore, G. C., I. Benbasat. 1991. Development of an instrument to measure the perceptions of adopting an

information technology innovation. Information Systems Research, 2(3), 192-222.

Nerney, C. 2014. The Android one could gut the market for high-end Android phones. September 10.

Accessed June 11, 2015. http://www.citeworld.com/article/2604668/mobile-byod/the-android-one-could-gut-

the-market-for-high-end-android-phones.html.

Noble, P. M., T. S. Gruca. 1999. Industrial pricing: Theory and practice. Marketing Science, 18(3), 435-454.

Parker, G., M. Van Alstyne. 2005. Two-sided network effects: A theory of information product design.

Management Science, 51(10), 1494-1504.

Porter, M. E. 1980. Competitive strategy: Techniques for analyzing industries and competitors, New York,

NY: Free Press.

Raudenbush, S. W., A. S. Bryk. 2002. Hierarchical linear models: Applications and data analysis methods.

Thousand Oaks, CA: Sage Publications Inc.

Rochet J. C., J. Tirole. 2003. Platform competition in two sided markets. Journal of the European Economic

Association, 1(4), 990-1029.

Rogers, E. M. 1983. Diffusion of Innovations, New York, NY: Free Press.

Rosenbaum, P., D. Rubin. 1983. The central role of the propensity score in observational studies for causal

effects. Biometrika, 70(1), 41-55.

Rosenbaum, P., D. Rubin. 1984. Reducing bias in observational studies using subclassification on the

propensity score. Journal of the American Statistical Association, 79(387), 516-524.

Shapiro, C., H. Varian. 1999. Information rules: A strategic guide to the network economy. Cambridge, MA:

Harvard Business School Press.

32

Siegfried, J. J., L. B. Evans. 1994. Empirical studies of entry and exit: A survey of the evidence. Review of

Industrial Organization, 9(2), 121-155.

Spann, M., M. Fischer, G. J. Tellis. 2015. Skimming or penetration? Strategic dynamic pricing for new

products. Marketing Science, 32(2), 235-249.

Tellis, G. J. 1986. Beyond the many faces of price: An integration of pricing strategies. Journal of

Marketing, 50(4), 146-160.

�!%�����������1(0��������4,!.%!*�!� %*"+.)�0%+*�#++ /��7�!./%+*-to-1,#.� !8���!�%/%+*��1,,+.0��5/0!)/��

56, 494-501.

Wilcox, M. 2014. Which app store should you use?. Vision Mobile. Accessed June 11, 2015.

http://www.developereconomics.com/which-app-stores-should-you-use/.

Williams, J. R. 2009. Mobile computing: predictions on sustainable advantage. Research Showcase Paper

541, Tepper School of Business, Carnegie Mellon University.

Yao, Y., J. Zhang. 2012. Pricing for shipping services of online retailers: Analytical and empirical

approaches. Decision Support Systems, 53(2), 368-380.

33

Table 1. Descriptive Statistics

Panel A : Descriptive statistics for categorical (dummy) variables

Mean (%) Mean (%)

Variables: Total App Store Google Play Variables: Total App Store Google Play

Store 50.00% 0.00% 100.00% Games 38.20% 46.70% 29.70%

Free trial version 15.63% 8.50% 22.77% Healthcare & F itness 2.83% 3.67% 2.00%

Two-sided market 8.98% 8.52% 9.45% Money & F inance 0.50% 0.00% 1.00%

Free version 31.86% 24.20% 39.52% Music 5.85% 6.17% 5.53%

In-app purchase 35.91% 44.15% 27.67% News & Info 0.42% 0.83% 0.00%

No app rating 0.96% 1.92% 0.00% Photo & Video 7.57% 11.37% 3.77%

Low app rating 2.43% 4.85% 0.00% Social Network 2.10% 2.17% 2.03%

Medium app rating 12.93% 18.23% 7.63% Themes & Customization 7.73% 5.57% 9.88%

High app rating 83.68% 75.00% 92.37% Travel 3.89% 2.75% 5.03%

Developer Type 86.03% 84.23% 87.83% TV 0.65% 1.30% 0.00%

Developer fame 22.37% 24.70% 20.04% Utility 21.52% 9.38% 33.66%

Education 1.98% 2.23% 1.72% Weather 3.09% 2.32% 3.87%

Entertainment 3.68% 5.55% 1.80%

Panel B : Descriptive statistics for continuous/integer variables

Total sample App Store Google Play

Mean Std Dev Min Max Mean Std Dev Min Max Mean Std Dev Min Max App size (Mbyte) 81.59 252.82 0.02 1945.6 108.27 265.56 0.2 1730.6 54.92 236.42 0.02 1945.6

Turnover 15.82 10.92 2 45 25.70 6.13 16 45 5.94 2.42 2 11 Number developer apps 23.73 55.26 1 480 29.90 55.02 1 354 17.57 54.81 1 480

Substitutes 518.54 623.20 0 1665 687.19 743.12 0 1665 349.86 409.45 0 1104 Time since release 575.98 423.14 1 1770 503.40 442.13 2 1770 648.57 390.00 1 1428

Price �:� 2.64 4.05 0.39 49.99 1.87 3.52 0.89 49.99 3.41 4.38 0.39 49.99

34

Table 2. Cor relation matrix

Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15)

(1) Developer Type

(2) Developer fame 0.22

(3) Number developer 0.20 0.51

(4) Free trial version 0.08 -0.04 -0.05

(5) Free version 0.02 -0.26 0.00 -0.29

(6) In-app purchase 0.23 0.26 0.26 -0.13 -0.05

(7) Two-sided market -0.03 -0.01 -0.01 -0.03 0.05 0.09

(8) Time since release 0.04 -0.08 -0.03 0.14 0.13 -0.14 0.09

(9) App size (Mbyte) 0.25 0.46 0.30 -0.12 -0.28 0.44 0.00 -0.25

(10) Substitutes 0.14 0.21 0.16 -0.12 -0.10 0.32 -0.01 -0.23 0.24

(11) No app rating -0.05 -0.04 -0.03 -0.04 -0.03 -0.05 -0.02 -0.09 -0.01 -0.01

(12) Low app rating -0.12 -0.04 -0.02 0.01 -0.05 -0.08 0.14 -0.10 -0.07 -0.06 -0.02

(13) Medium app rating -0.07 0.28 0.11 -0.11 -0.19 -0.04 0.07 0.03 0.11 -0.03 -0.04 -0.06

(14) High app rating 0.13 -0.22 -0.08 0.12 0.20 0.08 -0.12 0.04 -0.07 0.06 -0.22 -0.36 -0.87

(15) Store 0.05 -0.06 -0.16 0.20 0.16 -0.17 0.02 0.17 -0.43 -0.14 -0.10 -0.16 -0.16 0.24

(16) Turnover -0.05 0.04 0.14 -0.18 -0.15 0.16 -0.01 -0.16 0.39 0.12 0.10 0.15 0.14 -0.22 -0.92

Pearson correlation coefficients are shown to be significant (at 5% level) when above 0.03 in absolute value. Note also that sake of space we do not report Pearson correlation coefficients related to category dummies. At any rate, the correlation coefficient between Games and Substitutes (equal to 0.70) is the only relevant correlation coefficient involving category dummies, but this does not affect our results as we choose Games as the

baseline category.

Table 3. Means and relative differences before and after matching

Means before base matching Means after base matching

App Store Google Play Mean Difference App Store Google Play Mean Difference

Price 1.87 3.41 1.54*** 1.60 2.13 0.53***

Turnover 25.70 5.94 19.76*** 25.68 5.96 19.72***

Substitutes 687.19 349.86 337.33*** 1050.30 682.37 367.93***

Legend: + p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001

35

Table 4. Distr ibution platform effect and its interaction with developer decisions at the app level under the matched sample

(1) (2) (3) (4) (5) (6) (7) (8) Developer type 0.199** 0.257*** -0.00619 -0.0117 0.185** 0.196** 0.215** 0.198**

(0.0696) (0.0748) (0.00842) (0.00994) (0.0691) (0.0697) (0.0678) (0.0675) Developer fame 0.246** 0.259** 0.0208* 0.0108 0.243** 0.247** 0.218** 0.217**

(0.0783) (0.0855) (0.00973) (0.0111) (0.0777) (0.0785) (0.0757) (0.0756) Number developer apps (Ln) -0.0275 -0.0358+ -0.00334 0.00177 -0.0275 -0.0272 -0.0289 -0.0284

(0.0192) (0.0202) (0.00257) (0.00292) (0.0191) (0.0192) (0.0187) (0.0186) Free trial version 0.0428 0.0839 -0.0155** -0.0224*** -0.0499 0.0369 0.0520 -0.0363

(0.0488) (0.0510) (0.00579) (0.00615) (0.0576) (0.0490) (0.0482) (0.0575) Free version -0.0650 -0.0304 0.000421 -0.00572 -0.0360 -0.0644 -0.0906 -0.0649

(0.0580) (0.0608) (0.00548) (0.00607) (0.0581) (0.0579) (0.0569) (0.0572) In-app purchase -0.0259 -0.0338 -0.00139 -0.00370 -0.0247 -0.0243 -0.0138 -0.0114

(0.0521) (0.0541) (0.00450) (0.00487) (0.0514) (0.0520) (0.0510) (0.0504) Two-sided market -0.151 -0.0845 -0.00252 -0.0121+ -0.120 0.0146 -0.163+ 0.0681

(0.100) (0.101) (0.00619) (0.00632) (0.0995) (0.173) (0.0986) (0.169) Time since release 0.000246*** 0.000168** 0.00000672 0.0000161** 0.000248*** 0.000243*** 0.0000848 0.0000891

(0.0000560) (0.0000593) (0.00000585) (0.00000625) (0.0000554) (0.0000559) (0.0000666) (0.0000663) App size (Ln) 0.0391** 0.0198 0.000554 0.00409+ 0.0385** 0.0396** 0.0430** 0.0430**

(0.0148) (0.0146) (0.00196) (0.00217) (0.0147) (0.0148) (0.0146) (0.0146) Substitutes (Ln) -0.0585** -0.0756*** 0.0000464 0.00371 -0.0602** -0.0583** -0.0592** -0.0603**

(0.0209) (0.0223) (0.00254) (0.00306) (0.0207) (0.0209) (0.0204) (0.0202) High app rating 0.0706 0.0763+ 0.00595 0.00820 0.0692 0.0683 0.0751 0.0709

(0.0461) (0.0441) (0.00694) (0.00755) (0.0460) (0.0462) (0.0458) (0.0457) Store 0.232*** -0.0317*** 0.173** 0.243*** 0.209*** 0.174**

(0.0521) (0.00650) (0.0553) (0.0530) (0.0508) (0.0547) Turnover (Ln) -0.0345*** -0.00354

(0.00767) (0.00233) Free trial version x Store 0.305** 0.259*

(0.102) (0.101) Two-sided market x Store -0.246 -0.300

(0.210) (0.205) Time since release x Store 0.000436*** 0.000420***

(0.000102) (0.000102) App Category dummies YES YES YES YES YES YES YES YES

N 3520 3124 1446 1283 3520 3520 3520 3520 Developer-Store obs 262 244 126 117 262 262 262 262

App-Store obs 347 322 169 158 347 347 347 347 Wald test - pvalue 0.0000 0.0000 0.0000 0.0076 0.0000 0.0000 0.0000 0.0000

Legend: + p < 0.10, * p < 0.05, ** p < 0.01, *** p < 0.001. Standard errors in parentheses. Note that the constant and the app-store and developer-store pair random effects are always strongly significant. Also, No app rating and Low Rating dummies are not reported as they display no observations in the matched sample.